US7085685B2ExpiredUtilityPatentIndex 65
Device and method for filtering electrical signals, in particular acoustic signals
Est. expiryAug 30, 2022(expired)· nominal 20-yr term from priority
H04R 3/005H04R 25/407H04R 25/507H04R 2225/41H04R 2410/05
65
PatentIndex Score
8
Cited by
14
References
37
Claims
Abstract
A device for filtering electrical signals has a number of inputs arranged spatially at a distance from one another and supplying respective pluralities of input signal samples. A number of signal processing channels, each formed by a neuro-fuzzy filter, receive a respective plurality of input signal samples and generate a respective plurality of reconstructed samples. An adder receives the pluralities of reconstructed samples and adds them up, supplying a plurality of filtered signal samples. In this way, noise components are shorted. When activated by an acoustic scenario change recognition unit, a training unit calculates the weights of the neuro-fuzzy filters, optimizing them with respect to the existing noise.
Claims
exact text as granted — not AI-modified1. A device to filter electrical signals, having a number of input terminals arranged spatially at a distance from one another to supply respective pluralities of input signal samples, and a device output terminal to supply a plurality of filtered signal samples, the device comprising:
a number of signal processing channels, each signal processing channel being formed by a neuro-fuzzy filter to receive a respective plurality of input signal samples and to generate a respective plurality of reconstructed samples;
an adder unit to receive said plurality of reconstructed samples and having an output terminal to supply said plurality of filtered signal samples; and
routing means coupled to said output terminal of said adder unit and controllable so as first to supply said filtered signal samples back to said signal processing channels, then to supply said filtered signal samples to said device output terminal, wherein each signal processing channel includes;
a sample input terminal to receive alternately said input signal samples and said filtered signal samples and to supply signal samples to be filtered;
a signal feature computing unit to receive a respective plurality of samples to be filtered and to generate signal features;
a neuro-fuzzy network to receive said signal features and to generate reconstruction weights; and
a signal reconstruction unit to receive said samples to be filtered and said reconstruction weights and to generate said reconstructed samples from said samples to be filtered and said reconstruction weights.
2. The device according to claim 1 wherein said signal feature computing unit generates, for each said sample to be filtered:
a first signal feature correlated with a position of a sample to be filtered within an operative sample window;
a second signal feature correlated to a difference between said sample to be filtered and a central sample within said operative sample window; and
a third signal feature correlated to a difference between said sample to be filtered and an average sample value within said operative sample window.
3. The device according to claim 1 , further comprising a current-weights memory connected to said neuro-fuzzy filters and to store filter weights.
4. The device according to claim 3 , further comprising a weight training unit to calculate in real time said filtering weights.
5. The device according to claim 4 wherein said weight training unit comprises:
a training signal supply unit to supply a training signal having a known noise component;
a weight supply unit to supply training weights;
a spatial filtering unit to receive said training signal and said training weights and to output a filtered training signal;
a processing unit to process said training signal and said filtered training signal and to generate a fitness value; and
a control unit to repeatedly control said weight training unit and repeatedly receive said fitness value, said control unit being coupled to store the training weights having best fitness value in said current-weights memory.
6. The device according to claim 5 wherein said training signal supply unit includes a noise sample memory to store a plurality of noise samples, and a number of adders, one for each input of said device, each adder being coupled to receive a respective plurality of input signal samples and said noise samples, and to output a respective plurality of training signals.
7. The device according to claim 6 , further comprising a switching unit having a number of changeover switch elements, one for each signal processing channel, each changeover switch element having a first input terminal coupled to a respective input terminal of the device, a second input terminal coupled to an output terminal of a respective adder, and an output terminal coupled to a respective signal processing channel.
8. The device according to claim 5 wherein said weight supply unit comprises a random number generator.
9. The device according to claim 6 wherein said processing unit comprises means for computing a fitness degree correlated to a signal-to-noise ratio between said filtered training signal and said noise samples.
10. The device according to claim 5 , further comprising a best-fitness memory to store a best-fitness value and a best-weights value, wherein said control unit comprises comparison means for comparing said fitness value supplied by said processing unit and said best-fitness value, and writing means for writing said best-fitness memory with said fitness value, and said best-weight memory with corresponding training weights, in case said fitness value supplied by said processing unit is better than said best-fitness value.
11. The device according to claim 3 , further comprising an acoustic scenario change recognition unit to receive said filtered signal samples.
12. The device according to claim 11 wherein said acoustic scenario change recognition unit includes:
a subband-splitting block to receive said filtered signal samples from said device output and to generate a plurality of sets of samples;
a features extraction unit to calculate features of each set of samples;
a neuro-fuzzy network to generate acoustically weighted samples; and
a scenario change decision unit to receive said acoustically weighted samples and to output an activation signal for activation of said weight training unit.
13. The device according to claim 12 wherein said subband splitting block includes a plurality of splitting stages in cascade.
14. The device according to claim 13 wherein each said splitting stage includes:
a first and a second filter, in quadrature to each other, to receive a stream of samples to be split and to generate each a respective stream of split samples; and
a first and a second downsampler unit, each to receive a respective said stream of split samples.
15. The device according to claim 14 wherein said first filter of said splitting stages is a lowpass filter, and said second filter of said splitting stages is a highpass filter.
16. The device according to claim 12 wherein said feature extraction unit calculates energy of each set of samples.
17. The device according to claim 12 wherein said neuro-fuzzy network comprises:
fuzzification neurons to receive said signal features, and to generate first-layer outputs that define a confidence level of said signal features with respect to membership functions of a triangular type;
fuzzy AND neurons to receive said first-layer outputs and to generate second-layer outputs derived from fuzzy rules; and
a defuzzification neuron to receive said second-layer outputs and to generate an acoustically weighted sample for each of said filtered samples, using a gravity-of-gravity criterion.
18. The device according to claim 12 wherein said scenario change decision unit generates said activation signal by digitization at least one of said acoustically weighted samples.
19. The device according to claim 17 , further comprising:
a clustering training network having a first input terminal to receive said filtered signal samples from said device output terminal, a second input terminal to receive said acoustically weighted samples, and an output terminal connected to the clustering weights memory, said clustering training network including:
energy calculation means for calculating a mean energy of said filtered signal samples in a preset operative window;
gravity-of-gravity calculating means for determining centers of gravity of said membership functions according to said mean energy, said gravity-of-gravity calculating means being coupled and supplying said centers of gravity to said fuzzification neurons;
random generator means for randomly generating weights for said second-layer and third-layer neurons;
fitness calculation means for calculating a fitness function from said filtered signal samples and target signal samples;
fitness comparison means for comparing said calculated fitness function with a previous stored value;
storage means for storing said fitness function, said centers of gravity and said weights, in case said calculated fitness function is better than said previous stored value; and
next-activation means for activating said energy calculation means, said gravity-of-gravity calculation means, said random generator means, said fitness comparison means, and said storage means.
20. A method for filtering electrical signals, comprising:
receiving a plurality of streams of signal samples to be filtered; and
generating a plurality of filtered signal samples, wherein said generating includes:
receiving alternately said signal samples to be filtered and feedback filtered signal samples, and supplying these signal samples for filtering;
obtaining signal features for the supplied signal samples;
filtering the supplied signal samples through a respective neuro-fuzzy filter that use the obtained signal features to generate reconstruction weights;
generating a plurality of streams of reconstructed samples based on the reconstruction weights; and
adding said plurality of streams of reconstructed samples to obtain added signal samples.
21. The method according to claim 20 , further comprising:
supplying said added signal samples to said neuro-fuzzy filters; and
repeating said filtering and adding to obtain said filtered signal samples and to output said filtered signal samples.
22. The method according to claim 20 , further comprising weight training including:
supplying a training signal having a known noise component;
supplying filtering weights to said neuro-fuzzy filters;
filtering said signal samples to be filtered, to obtain a training filtered signal;
calculating a current fitness value from said training filtered signal samples;
comparing said fitness value with a previous fitness value; and
storing said fitness value and said filtering weights if said current fitness value is better than said previous fitness value.
23. The method according to claim 22 wherein said supplying filtering weights comprises randomly generating said filtering weights.
24. The method according to claim 23 wherein said randomly generating said filtering weights, filtering, calculating a current fitness value, comparing, and storing are repeated a preset number of times.
25. The method according to claim 22 wherein said supplying a training signal comprises adding a plurality of noise samples to said filtered signal samples.
26. The method according to claim 22 , further comprising recognizing acoustic scenario changes in said filtered signal samples and activating said training.
27. The method according to claim 26 wherein said recognizing comprises:
splitting said filtered signal samples into a plurality of subbands;
filtering said subbands through clustering neuro-fuzzy filters to obtain an acoustically weighted signal; and
activating said training if said acoustically weighted signal has a preset value.
28. The method according to claim 27 wherein said splitting includes filtering said subbands using filters having a pass band correlated to bands that are critical for a human ear.
29. The method according to claim 26 , further comprising clustering training including:
calculating a mean energy of said filtered signal samples in a preset operative window;
determining centers of gravity of membership functions of said clustering neuro-fuzzy filters according to said mean energy;
calculating a fitness function from said filtered signal samples and target signal samples;
comparing said fitness function with a previous stored value; and
storing said fitness function and said centers of gravity, should said calculated fitness function be better than said previous stored value.
30. A system for filtering electrical signals, the system comprising:
means for receiving a plurality of streams of signal samples to be filtered; and
means for generating a plurality of filtered signal samples, including:
means for receiving alternately said signal samples to be filtered and feedback filtered signal samples, and for supplying these signal samples for filtering;
means for obtaining signal features for the supplied signal samples;
means for filtering the supplied signal samples through a respective neuro-fuzzy network that use the obtained signal features to generate reconstruction weights;
means for generating a plurality of streams of reconstructed samples based on the reconstruction weights; and
means for adding said plurality of streams of reconstructed samples to obtain added signal samples.
31. The system of claim 30 , further comprising means for updating filter weights used by the neuro-fuzzy network.
32. The system of claim 30 , further comprising means for detecting changes in an acoustic scenario.
33. The system of claim 32 , further comprising means for training the means for detecting changes in the acoustic scenario.
34. The method of claim 20 wherein the plurality of streams of signal samples to be filtered are derived from signals received by a plurality of sensors arranged symmetrically relative to a source of the signals.
35. The device according to claim 1 wherein the reconstructed samples generated by the signal reconstruction unit are calculated using equations:
oL
(
i
)
=
∑
j
=
0
2
N
oL3L
(
i
-
j
)
·
eL
(
i
-
j
)
∑
j
=
0
2
N
eL
(
i
-
j
)
and
oR
(
i
)
=
∑
j
=
0
2
N
oL3R
(
i
-
j
)
·
eR
(
i
-
j
)
∑
j
=
0
2
N
eR
(
i
-
j
)
,
wherein
:
oL(i), oR(i) are the reconstructed samples;
oL 3 L(i), oL 3 R(i) are the reconstruction weights;
eL(i), eR(i) are the samples to be filtered; and
N is a position of a central sample in a work window.
36. A device to filter electrical signals, having a number of input terminals arranged spatially at a distance from one another to supply respective pluralities of input signal samples, and a device output terminal to supply a plurality of filtered signal samples, the device comprising:
a number of signal processing channels, each signal processing channel being formed by a neuro-fuzzy filter to receive a respective plurality of input signal samples and to generate a respective plurality of reconstructed samples;
an adder unit to receive said plurality of reconstructed samples and having an output terminal to supply said plurality of filtered signal samples; and
at least one routing device coupled to said output terminal of said adder unit and controllable so as first to supply said filtered signal samples back to said signal processing channels, then to supply said filtered signal samples to said device output terminal, wherein each signal processing channel includes a signal feature computing unit to receive a respective plurality of samples to be filtered and to generate signal features, wherein the signal feature computing unit generates for each of said samples to be filtered:
a first signal feature correlated with a position of a sample to be filtered within an operative sample window;
a second signal feature correlated to a difference between said sample to be filtered and a central sample within said operative sample window; and
a third signal feature correlated to a difference between said sample to be filtered and an average sample value within said operative sample window.
37. The device according to claim 36 , further comprising a signal reconstruction unit in each of the signal processing channels to receive said samples to be filtered and to receive reconstruction weights, and to generate said reconstructed samples from said samples to be filtered and said reconstruction weights, wherein the reconstructed samples generated by the signal reconstruction unit are calculated using equations:
oL
(
i
)
=
∑
j
=
0
2
N
oL3L
(
i
-
j
)
·
eL
(
i
-
j
)
∑
j
=
0
2
N
eL
(
i
-
j
)
and
oR
(
i
)
=
∑
j
=
0
2
N
oL3R
(
i
-
j
)
·
eR
(
i
-
j
)
∑
j
=
0
2
N
eR
(
i
-
j
)
,
wherein
:
oL(i), oR(i) are the reconstructed samples;
oL 3 L(i), oL 3 R(i) are the reconstruction weights;
eL(i), eR(i) are the samples to be filtered; and
N is a position of the central sample in the operative sample window.Cited by (0)
No later patents cite this yet.
References (0)
No backward citations on record.